- From: Nigel Megitt <nigel.megitt@bbc.co.uk>
- Date: Thu, 28 Sep 2023 16:24:47 +0000
- To: TTWG <public-tt@w3.org>
- Message-ID: <13EC6217-9416-4842-A6EF-C20F6010AA37@bbc.co.uk>
Thanks all for attending today’s TTWG meeting. Minutes can be found in HTML format at https://www.w3.org/2023/09/28-tt-minutes.html
Please note that we set a timeline of end of October to get IMSC-HRM into a state ready for requesting transition to CR.
Those minutes in text format:
[1]W3C
[1] https://www.w3.org/
Timed Text Working Group Teleconference
28 September 2023
[2]Previous meeting. [3]Agenda. [4]IRC log.
[2] https://www.w3.org/2023/09/12-tt-minutes.html
[3] https://github.com/w3c/ttwg/issues/263
[4] https://www.w3.org/2023/09/28-tt-irc
Attendees
Present
Atsushi, Cyril, Gary, Nigel, Pierre
Regrets
Andreas
Chair
Gary, Nigel
Scribe
nigel
Contents
1. [5]This meeting
1. [6]IMSC-HRM review feedback
2. [7]Agenda continued
2. [8]TPAC 2023 reflections
3. [9]IMSC-HRM
4. [10]DAPT
5. [11]Meeting close
Meeting minutes
This meeting
Nigel: Today our agenda is:
… TPAC 2023 reflections
… IMSC-HRM - do we have anything to cover on this today?
IMSC-HRM review feedback
Pierre: We received feedback from one content provider, which
was positive.
… I am working with three others.
… So far, it's been pretty good.
… Bugs were found in the reference implementation, which were
fixed.
… There are also bugs in some content providers' libraries.
… I'm cautiously optimistic that there are no major issues with
the spec itself.
… Now everybody is back from IBC and vacation I'm going to try
to get it done by end of October.
Cyril: You said one provider - that's Netflix, right?
Pierre: Yes, the only one who has provided feedback to the
group.
… One of the three others has provided me with some results
privately.
Cyril: I don't know what we decided - when we did the test
Nigel there were bugs, but that's okay.
… The Netflix content did not invalidate the IMSC-HRM model, so
it is probably good.
… One thing we found that was interesting, but still does not
jeopardise the model,
… for some content, Netflix produces content with very small
cues - I ran about 3000 pieces of content -
… and we decided the content should have been authored
differently.
… The issue is with 2 speakers speaking almost at the same
time, they have cues that overlap,
… but not completely, in time.
… If you have 2 speakers, one speaks, then the other starts
speaking, then the first stops,
… Netflix splits that into 3 non-overlapping ISDs. If they are
very short, that was creating a
… content validation error according to the HRM.
… I suggest we keep that as an issue and keep talking about it.
Pierre: Yes, I think that's worth discussing.
… The bottom line is that the HRM model does not assume that
the renderer can detect identical
… regions or parts of ISDs.
Nigel: It does assume some level of caching, at least.
Pierre: Yes but it has no notion of identical content, so
background redraws are not cached, for example.
… I'm not sure it's a problem.
… The Netflix approach, which Cyril will raise as an issue, was
introduced to work around some
… limitations of some clients.
Nigel: I think we just completed that agenda topic!
Agenda continued
Nigel: We may also have a few things to discuss on DAPT.
… In AOB, Andreas sent an email reminder about the DVB liaison,
and I have responded
… on the member-tt list. Not sure if we have anything more to
discuss during this call?
group: no request to discuss further today
Nigel: Is there any other business, or points to make sure we
cover?
Pierre: What's the plan wrt that liaison?
Nigel: Let's cover in TPAC 2023 reflections
TPAC 2023 reflections
Nigel: Just want to open up to any thoughts or observations
anyone had?
… I should comment about the joint meeting with APA WG and MEIG
on the Thursday afternoon.
… We had a good discussion, and it included the liaison from
DVB, which APA WG was interested in also.
… Since the liaison wasn't clearly targeted at any one group,
but several might be interested,
… I asked for a single team contact to bring together the
responses and look after sending them.
… François (tidoust) offered to do this or identify someone
else who would.
… In that context I think we should bring him in on the
reflector discussions.
… Does anyone have any comments on the response draft from
Andreas and my reply?
Pierre: Yes, sounds like a good idea to refresh our collective
memory.
[12]DVB Liaison (member only reflector link)
[12] https://lists.w3.org/Archives/Member/member-tt/2023Sep/0002.html
Nigel: [iterates through the liaison input]
Pierre: On the point of audio track language, and the
interaction between the audio signalled in the
… content and user preference, I think any metadata document
that describes what to signal in the content
… is not useful unless there's an algorithm indicating how that
metadata is used by the client.
… There are subtle interactions between choice of language on
the client, whether or not the client
… has indicated it would like captions or subtitles, and there
have been attempts at doing that.
… it's a surprisingly really hard problem.
… It would be great to standardise something, and the algorithm
should be standardised to, for the client.
Nigel: I would like to say that too.
Pierre: We should say it plainly: unless there's an algorithm
that specifies the interaction between
… user choices and signalling then it is incomplete work.
… There was an algorithm created back in the Ultraviolet/DECE
days, and it's quite complex.
Nigel: Yes, I tried reading that once!
Pierre: We should really emphasise that point.
… It's particularly important when forced narrative is
available.
Nigel: Yes
… Any more on that, or on TPAC more generally?
Pierre: On TPAC more generally, I was there for the discussion
about Apple's suggestion on an API
… for improving TextTrackCue. Do you know if this is going to
be turned into an effort within this
… group or another group?
Gary: I think it's on Apple to take the action of updating all
the relevant specs like HTML, WebVTT etc
[13]WHATWG pull request on HTML
[13] https://github.com/whatwg/html/pull/9771
Nigel: They have opened the above pull request on HTML.
… There's a lot of HTML spec complexity in there, but in terms
of the basic requirements I have
… added a couple of comments, and recommend others review too.
Nigel: The issue I have with this now is that without full CSS
support for fillLineGap and linePadding I don't know how we
could use it.
Pierre: In creating imscJS we spent a lot of time working out
how to make it match the TTML expectations.
… I'm not aware of any effort being done to validate this API,
is my summary.
Gary: From my understanding, the API doesn't preclude any of
that, the only change in IMSC is to add
… the attribute tags to the output HTML from imscJS, so the
styling should just be carried over.
Pierre: I can believe that theoretically, but we need to see a
comparison between the rendered output and
… the test references.
Gary: I guess that's part of getting the pull request approved.
Pierre: The reason I'm raising it is that HTML and CSS and TTML
are each complex.
… I've surprised myself in the past with corner cases.
Gary: They brought up that there can be conflicts between the
user setting overrides and the
… IMSC presentation. That's another issue - they said there's
nothing they can do about that.
Nigel: Two things. Firstly, without CSS support for fillLineGap
and linePadding, I think it will be impossible
… to make the tests all render correctly, because the imscJS
code that works around those HTML and CSS
… limitations cannot run without the HTML fragment being homed
to some DOM element.
… And in the proposal, no client code can run on it when that
is the case.
… Secondly, if CSS support were added for those features, then
that could work around some of the
… oddities that could result from user settings being
unexpected, in respect of those two styling features.
… I haven't even thought about ruby and text decoration but I
think that ought to work, in principle.
… So maybe the feedback to Apple is yes, but we need CSS
support for those missing features.
Gary: Yes, and I think that was discussed and maybe we need to
bring that in more strongly that as
… part of this the CSS related discussions need to be
restarted.
… To the first point, potentially, and I guess this is the
question, the document fragment we give to the API,
… there's nothing disallowing it from being added to the DOM
first to apply the workaround beforehand.
… Theoretically this could be done, as a workaround.
Nigel: Yes, though if the user makes the text size bigger then
it would break.
Pierre: I don't think anyone has implemented imscJS this way
and tested it fully.
Gary: I think Eric did implement it this way, but maybe did not
cover all the edge cases.
Pierre: This API is definitely a bit different from what I saw
4 or 5 years ago.
Gary: I don't disagree that a lot more testing is needed,
especially with more complex inputs.
Nigel: Sounds like an action to land this point somehow, not
sure who is best placed to do so.
… I'm certainly happy to send a message to Eric and James from
Apple, as well as Marcos.
Gary: Final point: they link to some tests they wrote for
WebKit in their PR, but it does seem to be on the simple side.
Nigel: Any other TPAC discussion points?
group: none
IMSC-HRM
Cyril: What are the next steps to move forward?
… One more content provider reporting and that is it?
Nigel: Good question - what are the exit criteria?
… We have tests
Pierre: Yes.
… I think we need 2 content providers. We have one.
… I'm hoping that we can, in a month, decide whether or not we
need changes.
Nigel: [reads CR exit criteria at [14]https://www.w3.org/TR/
imsc-hrm/#sotd]
[14] https://www.w3.org/TR/imsc-hrm/#sotd]
Cyril: So we already meet the criteria?
Nigel: Apparently so, though in the weakest way we possibly
could!
Atsushi: I'm not sure what the first criteria means - do we
need a content document
… produced by a content producing implementation and validated
by a content validator,
… but I am not sure. We have a manual set of test suites and
validated by [scribe missed]
… but if current criteria are that the same document needs to
be produced by one implementation
… and validated separately, then we need some other set of
content.
Nigel: We discussed at the top of the meeting, and mentioned
that Netflix has an implementation
… that they have verified by processing about 3,300 documents
through the validator.
Atsushi: That's great!
Cyril: I did send an email. There were ~20 languages, some
subtitles for deaf and hard of hearing,
… some forced narratives, some translation subtitles.
Pierre: As another data point:
<pal> From another content platform: I ran the tool over the
weekend on 25,000 randomly selected samples from our library. I
recorded 100 failures and I have attached the output of the
tool for those failures.
Pierre: I'm trying to get them to release those results.
… So far I think all the failures are in the files themselves.
That gives a sense of the scale.
… 25,000 TTML files.
… So far all the failures were errors in the TTML that probably
came from errors in translation from 608,
… by my guess.
Nigel: Syntactical errors, or something like that?
Pierre: Not sure what you'd call them - for example, timestamps
in hh:mm:ss:ff and the frame counter
… goes beyond the frame rate, e.g. if 30fps, and a frame count
of 30!
Nigel: Occasionally we see errors like that too in our tooling
in the BBC, which we do catch.
Pierre: We are trying to complete the CR exit criteria report
by the end of October.
Nigel: Yes, good idea, let's try to get this completed soon -
does that timescale work for everyone?
… I'm going to record assent by silence here!
… That's great, gives us about a month to verify that we have
met the exit criteria.
DAPT
Nigel: I have one question - anything from you Cyril?
Cyril: Wide review inputs
Nigel: Yes, I have been making the point generally, on email
and to people at IBC, that now is the time
… to review the spec and provide comments, while we're in WD
and more easily can make changes.
… I did talk to 3 or 4 organisations about DAPT specifically
and some were very positive, and said they
… would either be implementing or reviewing or both.
… It was extremely positive.
Cyril: I did open the TAG review and the i18n review, since we
last talked.
… I haven't received any feedback yet, though it has not been
long.
Atsushi: For i18n we just reviewed it and resolved it as
completed, for information.
… We will mark it as completed shortly.
Cyril: Great!
Atsushi: The action on GitHub might take a little time.
Cyril: I can see that aphilips moved it from in-review to
completed an hour ago.
Nigel: One question from me: I thought we had resolved to make
langSrc be absent or a language code,
… but couldn't find it documented.
Cyril: Yes we did
Atsushi: It's in our minutes
Gary: You're not misremembering.
Nigel: Thank you! I think the action is on me to implement
that, so I will go ahead with that.
… The other thing I wanted to note was that I just opened an
editorial pull request in response to an
… issue raised by Andreas, about the definitions of script and
transcript, so if anyone can review that,
… it's only small, and that'd be helpful!
<atsushi> [15]https://www.w3.org/2023/09/
12-tt-minutes.html#x959 ?
[15] https://www.w3.org/2023/09/12-tt-minutes.html#x959
[16]Redraft opening section of §2.1 w3c/dapt#183
[16] https://github.com/w3c/dapt/pull/183
Cyril: Did you have any feedback from privacy and security
reviews?
Nigel: No not yet
Cyril: We discussed removing styles too?
Nigel: Yes we agreed to do that, in [17]w3c/dapt#124
[17] https://github.com/w3c/dapt/issues/124
Cyril: The language one was [18]w3c/dapt#148
[18] https://github.com/w3c/dapt/issues/148
Nigel: Ah, thank you, I will do that.
Meeting close
Nigel: Thank you everyone, we're at time, and just completed
the agenda. Let's adjourn. Be free!
… [adjourns meeting]
Minutes manually created (not a transcript), formatted by
[19]scribe.perl version 221 (Fri Jul 21 14:01:30 2023 UTC).
[19] https://w3c.github.io/scribe2/scribedoc.html
Received on Thursday, 28 September 2023 16:25:02 UTC